Kathy
@kathaem.bsky.social
Computational Linguistics / Multilingual Language Models
Into SciFi, choir, cats (incomplete list of interests)
they/them
Into SciFi, choir, cats (incomplete list of interests)
they/them
Idk about "primarily" mate
August 17, 2025 at 5:45 PM
Idk about "primarily" mate
You mean the most popular *US* politicians on this list
August 6, 2025 at 6:58 AM
You mean the most popular *US* politicians on this list
Personally, sleeping more and vitamin D in the winter.
...sorry, not much of a baker
...sorry, not much of a baker
July 27, 2025 at 9:04 PM
Personally, sleeping more and vitamin D in the winter.
...sorry, not much of a baker
...sorry, not much of a baker
As a second language English speaker this also confused me for so long. Eventually I decided it must be from the phrase "having cake" which also means eating the cake
April 6, 2025 at 9:45 AM
As a second language English speaker this also confused me for so long. Eventually I decided it must be from the phrase "having cake" which also means eating the cake
Oh very nice to see a paper for this intuition, and the data could be very useful! Adding to the reading list 👀
March 22, 2025 at 9:10 AM
Oh very nice to see a paper for this intuition, and the data could be very useful! Adding to the reading list 👀
Alignability is more predictive of cross-lingual transfer than divergence of literal token distributions, particularly for language pairs with disparate scripts.
March 3, 2025 at 5:04 PM
Alignability is more predictive of cross-lingual transfer than divergence of literal token distributions, particularly for language pairs with disparate scripts.
Basically we argue that token overlap measures for predicting multilingual performance are too literal, and introduce the notion of **token alignability**, which can be measured via the scores of a statistical aligner over a corpus tokenised with a given tokenised.
March 3, 2025 at 5:04 PM
Basically we argue that token overlap measures for predicting multilingual performance are too literal, and introduce the notion of **token alignability**, which can be measured via the scores of a statistical aligner over a corpus tokenised with a given tokenised.
Gotta say I'm not sure what pronunciation "luh-BOEV" is referring to but in my head it sounds like French beef
December 26, 2024 at 8:10 AM
Gotta say I'm not sure what pronunciation "luh-BOEV" is referring to but in my head it sounds like French beef
Germany. a) ground floor b) first floor. This matches how we count in German but the German terms basically treat the "upper floors" separately from the "ground floor"
December 18, 2024 at 9:17 AM
Germany. a) ground floor b) first floor. This matches how we count in German but the German terms basically treat the "upper floors" separately from the "ground floor"
5k is a small town, honestly 😂
November 20, 2024 at 12:50 PM
5k is a small town, honestly 😂
Just wanted to say a quick thank you for organising a lovely social! 🎊🌈
November 18, 2024 at 2:49 PM
Just wanted to say a quick thank you for organising a lovely social! 🎊🌈